diff --git a/unsloth/models/loader.py b/unsloth/models/loader.py
index 97add13f2d..9d2d3d9b06 100644
--- a/unsloth/models/loader.py
+++ b/unsloth/models/loader.py
@@ -100,6 +100,7 @@ global FORCE_FLOAT32
 # Forces float32 precision since float16 goes to infinity
 FORCE_FLOAT32 = [
     "gemma3,",  # Add comma bc gemma3 will match gemma3n
+    "gemma3text",  # Gemma3TextModel (EmbeddingGemma, standalone text-only Gemma3)
     "gemma3n",
     "gpt_oss",
 ]
@@ -116,6 +117,7 @@ global DISABLE_SDPA_MODEL_NAMES
 # Disables some SDPA modules since it's wrong
 DISABLE_SDPA_MODEL_NAMES = [
     "gemma3,",  # Add comma bc gemma3 will match gemma3n
+    "gemma3_text",  # Gemma3TextModel (EmbeddingGemma) - substring match, keep underscore
 ]
 
 
diff --git a/unsloth/models/sentence_transformer.py b/unsloth/models/sentence_transformer.py
index a3ea950420..ad59165a50 100644
--- a/unsloth/models/sentence_transformer.py
+++ b/unsloth/models/sentence_transformer.py
@@ -2089,6 +2089,20 @@ def _patch_sentence_transformer_trainer():
         # Call original __init__
         _original_init(self, *args, **kwargs)
 
+        # Disable mixed precision when FORCE_FLOAT32 is active (matches rl.py behavior)
+        if os.environ.get("UNSLOTH_FORCE_FLOAT32", "0") == "1":
+            if hasattr(self, "args") and self.args is not None:
+                if self.args.fp16 or self.args.bf16:
+                    print(
+                        "Unsloth: Switching to float32 training since model cannot work with float16"
+                    )
+                    self.args.fp16 = False
+                    self.args.bf16 = False
+                    if hasattr(self.args, "bf16_full_eval"):
+                        self.args.bf16_full_eval = False
+                    if hasattr(self.args, "fp16_full_eval"):
+                        self.args.fp16_full_eval = False
+
     SentenceTransformerTrainer.__init__ = _patched_init
     SentenceTransformerTrainer._unsloth_auto_compile_patched = True